Midjourney V7 vs Flux: How to Prompt Each Model in 2026

You can give Midjourney V7 and Flux the same brief and get two different pictures. That is not a bug. The models were trained to want different kinds of language.

Midjourney still rewards a short scene plus trailing flags. Flux rewards a complete description in ordinary sentences. If you paste a Discord prompt into a Flux box, the flags do nothing useful. If you paste a Flux paragraph into Midjourney, the end of the sentence often gets ignored.

This post is a conversion guide, not a winner-takes-all review. Use Midjourney when you want a strong house aesthetic and fast style lock. Use Flux when you need literal layout, readable text on the image, or an API you can script. Keep both prompt families in PromptDig (Browse more prompts) so you are not rewriting from memory every time.

Write the brief once, then fork it

Start with facts, not style words. Write a five-line brief a photographer could shoot:

• Subject: who or what, how many, what they are doing
• Setting: place, time of day, what is in the background
• Camera: distance, angle, lens feel (wide, 85mm portrait, overhead packshot)
• Light: direction, hardness, color
• Must include / must not include: logo, text, extra people, clutter

That brief is model-agnostic. Everything after this is translation.

For Midjourney V7, compress the brief into one or two sentences, then put control in parameters. Lead with the subject. Follow with lighting and materials. End with flags. V7 understands natural language better than older versions, so you do not need comma-stuffed tags. You still want the flags. They are how you lock canvas and taste.

A working V7 shape:

A ceramic pour-over dripper on a walnut counter, morning side light,
soft shadow to the left, linen towel in the background, documentary still
--ar 4:5 --stylize 150 --v 7

Use --stylize (or --s) when you want more or less of Midjourney's own look. Lower values stay closer to your words. Higher values look more "Midjourney." Use --sref with an image URL or a style code when the job is matching a brand mood, and --sw to turn that influence up or down. Use --oref when the job is keeping one product or person consistent across scenes, and --ow to weigh that reference. Use --no for things you want suppressed. Do not bury the "no" inside the prose.

For Flux, expand the same brief into grammatical sentences. Name positions. Flux is literal. If you do not say the towel is behind and to the right, it may invent a composition you did not ask for. Do not append --ar or --stylize. Set aspect ratio in the UI or API. Say what should be on the surfaces instead of listing what to avoid.

A working Flux shape of the same shot:

A single ceramic pour-over dripper sits on a walnut kitchen counter,
slightly off-center. Morning light comes from a window on the right,
leaving a soft shadow to the left. A folded linen towel rests behind
the dripper. No other objects on the counter. Photographed straight-on
from chest height, shallow depth of field, quiet documentary color.

Notice the extra spatial language. That is the real prompt skill for Flux, not a secret token.

What each model is actually good at

Midjourney V7 is strong when the brief is aesthetic. Campaign stills, moodboards, character sheets with --oref, and anything where "this should feel like our brand" matters more than "the label must be readable." Personalization and moodboards on midjourney.com are part of that workflow.

Flux is strong when the brief is a spec. Product shots with a known number of objects. UI mock textures. Posters and packaging where letters have to be letters. Anything you will generate in bulk through an API. Write the scene like you are directing a still photographer who has never seen your brand deck.

Text on the image is the usual fork. If the deliverable needs a headline on a book cover or a word on a mug, start in Flux. If the deliverable is atmosphere and you will composite type later, Midjourney is often faster.

Do not treat parameters as portable. --style raw, --chaos, --exp, and --v 7 are Midjourney-only. Guidance scale, steps, and pixel size are interface settings around Flux, not words you put in the prompt. When you convert a library, strip the flags first, then rewrite the scene.

A conversion checklist you can reuse

When a Midjourney prompt has to run on Flux:

1. Delete every -- flag.
2. Turn aspect ratio into a composition phrase or a UI setting ("vertical portrait, 4:5").
3. Replace mood adjectives with visible causes. "Ethereal" becomes "pale haze, backlight, low contrast."
4. Put objects in space: left, right, foreground, background, how many.
5. Describe the empty areas you care about ("clean counter, unmarked ceramic") instead of a negative list.

When a Flux paragraph has to run on Midjourney:

1. Cut to the first two sentences.
2. Move camera and canvas into --ar and a short lens phrase.
3. Move "do not" items into --no.
4. If you have a reference still, add --sref or --oref rather than describing the reference in prose.
5. Generate, then change one flag at a time. Do not rewrite the whole sentence because the crop was wrong.

Keep both versions of every prompt you actually ship. The pair is more valuable than a "universal" prompt that is mediocre on both models. If you want that pair to survive the next tool shuffle, share it on PromptDig (Share a prompt) with the model name in the title so people do not paste the wrong dialect.

You do not need a new personality for each model. You need a brief, then a fork. Write the facts once. Translate the control surface. That is the 2026 skill.